AI Training Is Copying. Is It Infringement? | Keith Kupferschmid, CEO Copyright Alliance

WEBVTT
1
00:00:00.031 --> 00:00:06.658
Welcome to Technically Creative, where we talk to creative leaders about how technology and AI are changing their industries and their companies.
2
00:00:06.859 --> 00:00:15.688
This episode is brought to you by the Future of Life Institute, an independent nonprofit working to steer transformative technologies like AI towards benefiting humanity.
3
00:00:15.708 --> 00:00:16.349
Thank you.
4
00:00:16.489 --> 00:00:20.551
Today's guest is Keith Kupferschmid, CEO of the Copyright Alliance.
5
00:00:21.111 --> 00:00:26.674
And I wanted to have this conversation because I think we've made copyright and AI almost impossibly confusing.
6
00:00:27.175 --> 00:00:33.038
We throw around terms like fair use, transformative use, training data, authorship, and infringement—
7
00:00:33.758 --> 00:00:36.080
Often as if we all agree on what they even mean.
8
00:00:36.420 --> 00:00:37.321
I've been guilty of it.
9
00:00:37.802 --> 00:00:42.826
And there was one distinction Keith made in this conversation that genuinely changed the way I understand the entire debate.
10
00:00:43.386 --> 00:00:47.230
We spend an enormous amount of time talking about what comes out of an AI model.
11
00:00:47.330 --> 00:00:50.773
But Keith argues that we keep in mind that there are two separate copyright questions.
12
00:00:50.813 --> 00:00:51.673
There's the output and there's the input.
13
00:00:51.693 --> 00:00:53.575
But Keith argues that we keep in mind that there are two separate copyright questions.
14
00:00:53.655 --> 00:00:57.601
Because before a model can generate anything, somebody had to put something into it.
15
00:00:58.162 --> 00:01:03.089
A book had to be copied, a song had to be ingested, an image had to be part of a data set.
16
00:01:03.369 --> 00:01:09.498
And whether the thing that eventually comes out looks anything like the original doesn't make that first question disappear.
17
00:01:10.139 --> 00:01:12.980
That sounds incredibly obvious once somebody explains it to you.
18
00:01:13.400 --> 00:01:17.202
But I realized during this conversation that I have been conflating these two things for years.
19
00:01:17.702 --> 00:01:22.585
And once you separate them, a lot of the AI copyright debate suddenly becomes much easier to understand.
20
00:01:23.105 --> 00:01:24.666
It also leads to a bigger question.
21
00:01:25.026 --> 00:01:29.368
I mean, AI isn't going away, and I don't think most artists even want it to.
22
00:01:29.728 --> 00:01:30.648
Some do, for sure.
23
00:01:31.128 --> 00:01:34.590
Creators have always adopted new tools, and many of them like using AI.
24
00:01:34.830 --> 00:01:38.892
So the interesting question isn't really whether AI and creativity can coexist.
25
00:01:39.352 --> 00:01:40.612
It's what the bargain should be.
26
00:01:40.752 --> 00:01:43.093
Copyright always had to serve two masters.
27
00:01:43.133 --> 00:01:47.075
It had to protect creators enough that making creative work remains valuable.
28
00:01:47.515 --> 00:01:52.077
But it also has to leave enough room for people to be inspired by what came before and create something new.
29
00:01:52.477 --> 00:01:54.038
So it has to be elastic like that.
30
00:01:54.758 --> 00:01:59.379
AI puts enormous pressure on the balance because it introduces something we've never really had before.
31
00:01:59.920 --> 00:02:07.022
The ability to ingest human creativity at an extraordinary scale and then produce competing material at an extraordinary scale.
32
00:02:07.622 --> 00:02:13.623
So how do we preserve the benefits of these tools without destroying the economic incentive to create the work they depend on?
33
00:02:13.823 --> 00:02:15.664
Keith has a surprisingly practical answer.
34
00:02:16.144 --> 00:02:16.705
Thank you.
35
00:02:38.567 --> 00:02:49.610
We get into the major AI copyright cases, fair use, what actually makes something copyrightable, why music may be particularly important to watch, and what creators should understand before using generative AI.
36
00:02:50.090 --> 00:02:55.411
And why it could still take years for the courts to resolve some of the most fundamental questions we're asking today.
37
00:02:56.111 --> 00:03:00.432
And Keith is very clear that protecting copyright doesn't mean being anti-AI.
38
00:03:01.052 --> 00:03:06.794
The question is whether we can build an AI economy where innovation and human creativity can actually thrive together.
39
00:03:06.914 --> 00:03:08.314
This is Keith Kufferschmidt.
40
00:03:16.781 --> 00:03:18.982
Keith Kupperschmidt, thank you so much for coming.
41
00:03:19.002 --> 00:03:21.044
You're the CEO of the Copyright Alliance.
42
00:03:21.544 --> 00:03:24.967
I feel like that is an increasingly important role to have.
43
00:03:25.748 --> 00:03:27.629
And I'm so excited to dig into this with you.
44
00:03:27.869 --> 00:03:29.150
Thanks for being on the podcast.
45
00:03:29.591 --> 00:03:30.712
Well, thank you for having me.
46
00:03:30.772 --> 00:03:32.133
I'm looking forward to it.
47
00:03:32.513 --> 00:03:33.173
Fantastic.
48
00:03:33.193 --> 00:03:34.974
Yeah, I'm looking forward to digging into this with you.
49
00:03:34.994 --> 00:03:40.917
Because, you know, people are throwing around the term copyright all over the place and fair use and this, that and the other thing.
50
00:03:40.937 --> 00:03:47.640
And so I'm really keen to talk to you about the issues at play in modern and a lot of the modern technologies we're seeing.
51
00:03:47.800 --> 00:03:58.625
And, you know, IP and artist rights and things like that, you know, and we have viewers across film, television, music for someone outside the copyright world or the IP world.
52
00:03:58.825 --> 00:04:04.029
What do they need to understand about what the Copyright Alliance is and who you represent?
53
00:04:04.290 --> 00:04:16.940
Yeah, so our mission ultimately is to protect individual creators and copyright owners and to make sure that the public and creators and copyright owners all understand the value and importance of copyright law.
54
00:04:17.341 --> 00:04:24.787
Copyright law for many hundreds of years, honestly, has served to protect and incentivize creativity.
55
00:04:25.647 --> 00:04:28.048
And and it continues to do so.
56
00:04:28.268 --> 00:04:39.072
And every once in a while, it is subject to some type of technology that challenges the copyright law, whether that's the photocopier, maybe or maybe it's the Internet.
57
00:04:39.232 --> 00:04:41.353
And most recently, artificial intelligence.
58
00:04:41.853 --> 00:04:45.875
And then we have to all sit back and go, OK, do we need to change the law?
59
00:04:45.895 --> 00:04:48.116
Is the law sufficient as as it stands?
60
00:04:49.456 --> 00:04:49.957
And look at that.
61
00:04:50.017 --> 00:04:53.558
And so the members that I represent, whether it's movie studios or.
62
00:04:54.138 --> 00:05:08.868
or record labels or book publishers or magazine and newspaper publishers or video game companies and so on and so forth, as well as the thousands of individual creators all rely on copyright law and they look to the Copyright Alliance to help them in that regard.
63
00:05:09.028 --> 00:05:11.891
It was so interesting that you brought up, you know, photocopiers.
64
00:05:12.331 --> 00:05:16.736
It feels like that's sort of what we're dealing with with AI, isn't it?
65
00:05:16.776 --> 00:05:24.343
It's just something that can copy and replicate ad infinitum at a breakneck speed.
66
00:05:24.424 --> 00:05:26.506
We've just changed the speed of it, really.
67
00:05:26.946 --> 00:05:28.287
Yeah, I mean, it's interesting.
68
00:05:28.847 --> 00:05:35.309
So this year is the 50th anniversary of the 1976 Copyright Act, the same Copyright Act that's in effect today.
69
00:05:35.970 --> 00:05:44.813
And what we did is we brought together what we call the founders, the people who wrote that copyright law, implemented that copyright law.
70
00:05:45.214 --> 00:05:46.614
We brought them together for a Zoom.
71
00:05:47.014 --> 00:05:48.215
And it was interesting that...
72
00:05:48.695 --> 00:05:52.618
You know, they brought up photocopying and AI in the same breath, sort of just like I did.
73
00:05:53.018 --> 00:05:58.341
And basically saying, look, each each generation thinks, oh, this is the biggest thing in the world universe.
74
00:05:58.381 --> 00:06:02.064
And and, you know, maybe we need to change copyright law because of that.
75
00:06:02.264 --> 00:06:03.125
But really, it's not.
76
00:06:03.145 --> 00:06:05.046
It's just another another challenge.
77
00:06:05.326 --> 00:06:13.550
I imagine that with any new technology, does your job change a lot or is it just kind of like, I'm always in the same old business.
78
00:06:13.610 --> 00:06:16.092
The world around me changes, but I'm always in the same old business.
79
00:06:16.752 --> 00:06:22.455
Or do you go, no, AI is fundamentally different and we have to respond to it differently?
80
00:06:22.715 --> 00:06:25.056
Yeah, AI is fundamentally different.
81
00:06:25.136 --> 00:06:42.001
It is an existential challenge because unlike other innovations and new technologies that come along, AI is different because what's happening here is you've got AI companies that are training their models by ingesting copyrighted works.
82
00:06:42.701 --> 00:06:54.777
to produce or to output works or whatever their output is called that will compete directly with those works that it was trained on and therefore put those creators out of business, if you will.
83
00:06:55.257 --> 00:06:59.383
That is a real, real existential challenge for the creative community.
84
00:06:59.663 --> 00:07:03.264
Yeah, it feels like it's a bit like the photocopier.
85
00:07:03.604 --> 00:07:06.705
Its ability to do it at scale is actually the scariest thing.
86
00:07:06.906 --> 00:07:21.170
It's one thing, and this is where I've always jarred with the fair use argument, that the fair use argument says, well, just like a human brain, you know, I'm allowed to input any number of anything from this wall of creativity behind me.
87
00:07:21.411 --> 00:07:22.191
I'm allowed to read it.
88
00:07:22.451 --> 00:07:23.671
I'm allowed to be inspired by it.
89
00:07:23.751 --> 00:07:28.073
I'm allowed to create something that is substantially different and it's allowed to be out there.
90
00:07:28.093 --> 00:07:28.633
Yeah.
91
00:07:28.733 --> 00:07:33.376
And there's that old adage that, you know, good artists borrow, great artists steal outright.
92
00:07:34.316 --> 00:07:41.100
You know, and I think that, however, the difference is my human brain only has a certain amount of output capacity.
93
00:07:41.640 --> 00:07:45.422
I can only compete by writing one or two.
94
00:07:45.442 --> 00:07:47.383
If I'm Stephen King, probably three novels a year.
95
00:07:47.403 --> 00:07:52.406
But, you know, there is there's a certain limitation there, whereas the limitation is.
96
00:07:53.366 --> 00:08:00.388
gates have been completely opened with AI, leading to incredible amounts of competition and I think disenfranchising artists.
97
00:08:00.428 --> 00:08:01.368
Is that a fair assessment?
98
00:08:01.668 --> 00:08:04.649
Yeah, you are 100% correct, but I want to point out something else.
99
00:08:04.669 --> 00:08:08.029
You talk about those books and other things that are on your shelves behind you.
100
00:08:08.650 --> 00:08:15.291
I bet you you paid for all those books and other doodads and trinkets that are up there.
101
00:08:15.311 --> 00:08:17.332
AI is not.
102
00:08:17.352 --> 00:08:19.612
Thanks for mentioning the doodads and trinkets.
103
00:08:20.752 --> 00:08:22.453
They're really what most of this background is
104
00:08:23.933 --> 00:08:25.274
Well, books, books too, right?
105
00:08:25.635 --> 00:08:27.376
Yeah, right, right.
106
00:08:27.396 --> 00:08:28.136
They're just for show.
107
00:08:28.797 --> 00:08:29.717
Yeah.
108
00:08:30.158 --> 00:08:33.260
And so, but you paid for those, but AI is not.
109
00:08:33.280 --> 00:08:45.508
Honestly, that is the crux of the problem is that AI is just, these AI companies are just taking and stealing without compensating the creators and the copyright owners and without getting their consent.
110
00:08:46.289 --> 00:08:49.130
And that's very different than the books on your shelves.
111
00:08:49.731 --> 00:08:52.313
And what makes it even worse is that not only are they...
112
00:08:52.973 --> 00:09:01.163
taking this stuff without permission and without compensation, they're then using it in a way that puts those very same people out of business.
113
00:09:01.304 --> 00:09:03.206
And so it's a double-edged sword.
114
00:09:03.226 --> 00:09:06.210
It's a real, real challenge for the creative community.
115
00:09:06.540 --> 00:09:09.202
And that's the line that was that was drawn in the sand.
116
00:09:09.222 --> 00:09:18.829
We had Lisa Oratz on not too long ago who, you know, she was part of the first case for critters that protected, you know, an A.I.
117
00:09:18.889 --> 00:09:21.130
output by showing meaningful human intervention.
118
00:09:21.191 --> 00:09:24.973
And and she's you know, she's a thought leader, just as you are on this space.
119
00:09:25.354 --> 00:09:31.438
But she kind of said that that was the interesting thing about the Barts versus Anthropic case was that they created a distinction between.
120
00:09:31.858 --> 00:09:37.603
Between books that had been purchased and that that was fair use, but books that had been pirated.
121
00:09:38.544 --> 00:09:44.769
That's where the demarcation line was for fair use training data.
122
00:09:45.049 --> 00:09:48.272
Yeah, I mean, to some extent, that's certainly what the Barts case said, right?
123
00:09:48.292 --> 00:09:50.994
The judge in the Barts case, who's now judge is now retired.
124
00:09:51.034 --> 00:09:55.218
But to be honest, if you that's in the Ninth Circuit, that case, the Barts case.
125
00:09:56.098 --> 00:09:58.980
The second circuit, which is the New York area, right?
126
00:09:59.240 --> 00:10:04.004
The second circuit took up similar issues, not in the AI context, but a similar issue.
127
00:10:04.024 --> 00:10:09.808
The idea of being able to buy books and then copy them and digitize them and use them as you would like.
128
00:10:10.368 --> 00:10:14.251
And the second circuit came out with a different, very, very different result.
129
00:10:14.271 --> 00:10:15.972
They said that is not their use.
130
00:10:15.992 --> 00:10:17.053
You cannot do that.
131
00:10:17.413 --> 00:10:24.018
And so you've got these two competing circuits, one on the West Coast and one on the East Coast, that came out with very, very different results.
132
00:10:24.538 --> 00:10:40.765
answers to that question and i think the east coast case did focus on that uh that um competition part of it a bit more is that right oh oh oh absolutely the fact that you know you first of all that you're digitizing something you're changing the format of it you you're not just
133
00:10:41.145 --> 00:10:50.688
taking a book, for instance, and giving it to a friend, which would be allowable under what's called the first sale doctrine, what you're doing is you're making a copy of that book or books, right?
134
00:10:51.408 --> 00:10:57.150
And you're – by digitizing it, and that's very different, and that is not specifically allowed by the copyright law.
135
00:10:57.490 --> 00:11:03.392
So then it forces people to go, okay, well, if it's not allowed by the copyright law, would it be allowed as a fair use?
136
00:11:03.872 --> 00:11:05.193
And the second circuit said no.
137
00:11:05.553 --> 00:11:08.814
The ninth circuit said in this particular instance, yes.
138
00:11:09.154 --> 00:11:29.083
And then, but to get to your point too, the ninth circuit in the BART's case also looked at the going to these pirate websites, which is pretty much what all the AI companies are doing, going to these pirate websites, they call them shadow libraries, but they're really not libraries at all, and scraping all this pirated material and using it to train their AI systems.
139
00:11:29.783 --> 00:11:35.106
And Joe Jalsup in the Barts case said, no, no, no, no, that is that's not above board.
140
00:11:35.986 --> 00:11:37.027
You're going to be liable for that.
141
00:11:37.047 --> 00:11:43.850
And they were in what was the biggest, I believe, at least the biggest damages case in copyright ever.
142
00:11:44.551 --> 00:11:50.434
And the authors are getting paid out, I think, somewhat handsomely in that case as a result.
143
00:11:50.775 --> 00:11:52.756
But it's all quite relative, though, isn't it?
144
00:11:52.796 --> 00:11:58.718
Because if there are, you know, a million authors that like, I mean, there's a lot of people who weren't party to that case.
145
00:11:59.138 --> 00:12:06.141
But then there were a lot of authors that were and then their proportionate share is is sort of negligible.
146
00:12:06.701 --> 00:12:09.983
Is the conversation moving so much past creators in this?
147
00:12:10.023 --> 00:12:10.363
I mean, like.
148
00:12:10.983 --> 00:12:17.330
Have the AI companies sort of won by guarding substantially the fair use argument?
149
00:12:17.590 --> 00:12:24.217
Is there much hope in these new cases or is the pendulum swinging very much in one, you know, inevitable direction?
150
00:12:24.578 --> 00:12:27.901
I see the pendulum is swinging toward the creative community.
151
00:12:27.921 --> 00:12:28.882
I don't think it initially was.
152
00:12:28.902 --> 00:12:29.243
Oh, great.
153
00:12:29.923 --> 00:12:32.565
I don't know what that will mean.
154
00:12:32.605 --> 00:12:38.129
And maybe this, and as we say in copyright, the pendulum tends to swing back and forth and back and forth, which is why we call it a pendulum.
155
00:12:38.509 --> 00:12:46.614
And so, although I think things are leading toward the correct decisions, ultimately we're going to have to see.
156
00:12:46.634 --> 00:12:48.375
There's over, I think, 120 people
157
00:12:49.496 --> 00:12:57.978
cases right now, as we talk today, that are out there pending, and they're pending all over the country, and there are different set of facts in all these cases.
158
00:12:58.378 --> 00:13:01.379
So we could get very different results in these cases.
159
00:13:01.759 --> 00:13:11.241
We could see the Ninth Circuit come out with some result that is very different than the Second Circuit or the Fifth Circuit, and we're going to have to make sense of all that.
160
00:13:11.681 --> 00:13:14.982
I mean, matter of fact, if I could just go back to talk about the Bartz case in the
161
00:13:19.744 --> 00:13:32.502
The cadre case came out in the same court, the same district court in the Ninth Circuit, Northern District of California, two different judges, and they came out with very different decisions, very, very different opinions, if you will.
162
00:13:32.902 --> 00:13:34.224
And so here you have –
163
00:13:35.045 --> 00:13:44.512
two judges in the same courtroom basically coming out with different opinions, you can imagine if you go across the country, you're going to get something even more different.
164
00:13:44.952 --> 00:14:00.483
And then in those cases involved text, you could be dealing with movies or music or images, and you're going to get vastly different opinions depending on what type of work the AI companies are copying and how they're copying it and where they're getting their source material from.
165
00:14:01.095 --> 00:14:01.416
Right.
166
00:14:01.616 --> 00:14:03.658
Because that was the very famous case, wasn't it?
167
00:14:03.758 --> 00:14:05.720
Sarah Silverman and George R.R.
168
00:14:05.760 --> 00:14:07.402
Martin, the writer of Game of Thrones.
169
00:14:07.902 --> 00:14:09.364
You know, that was at the beginning.
170
00:14:09.624 --> 00:14:12.327
That was the kind of the case that everybody was looking at.
171
00:14:12.387 --> 00:14:14.729
And actually, Bartz was heard first.
172
00:14:15.270 --> 00:14:17.612
It seemed the first win for fair use.
173
00:14:18.353 --> 00:14:18.513
Yeah.
174
00:14:19.214 --> 00:14:26.301
But like you said, you know, it seems like we expected these court cases to shed enormous amounts of lights on things.
175
00:14:26.321 --> 00:14:30.245
But it feels like, A, they're contradicting each other on fairly major points.
176
00:14:30.686 --> 00:14:34.490
And B, they raise a whole bunch more questions.
177
00:14:34.790 --> 00:14:39.314
Yeah, in some sense, it did confirm what we knew before the cases were decided, right?
178
00:14:39.714 --> 00:14:40.876
There was chaos before.
179
00:14:41.516 --> 00:14:43.358
We've confirmed there is chaos, right?
180
00:14:43.398 --> 00:14:46.461
Because you really look at these two cases and what are the takeaways?
181
00:14:46.961 --> 00:14:50.584
And I don't know that there's too many takeaways from either of those cases.
182
00:14:50.624 --> 00:14:50.925
I think –
183
00:14:52.025 --> 00:14:59.269
I think both sides will say, whether you're on the AI side or the creative community side, will say that, okay, the judge got this right but got this wrong.
184
00:15:00.630 --> 00:15:12.916
I think in the Bart's case in particular, Judge Alsop seemed to be so enamored with AI, the technology, that he said, of course, this is the most transformative technology of our time.
185
00:15:13.436 --> 00:15:14.437
And that's well and good.
186
00:15:14.617 --> 00:15:16.738
And it is amazing technology.
187
00:15:17.390 --> 00:15:21.172
But that is not the test for what is transformative and what is not transformative.
188
00:15:21.192 --> 00:15:22.152
It's actually a legal test.
189
00:15:22.572 --> 00:15:23.933
And he never applied the legal test.
190
00:15:23.953 --> 00:15:26.574
He just thought, oh, my God, this is so amazing technology.
191
00:15:26.614 --> 00:15:27.674
It must be transformative.
192
00:15:27.934 --> 00:15:30.575
But it's actually a legal test to figure out whether it's transformative.
193
00:15:30.595 --> 00:15:39.279
And I should let me take a step back and say that whether something is transformative or not is one element where you consider if something is fair use.
194
00:15:39.459 --> 00:15:41.019
There's lots of different factors to consider.
195
00:15:41.480 --> 00:15:44.561
Transformative use is sort of a sub factor, if you will.
196
00:15:44.841 --> 00:15:53.347
But even the transformation argument, I think that like what's so interesting is that I could write something, put it into AI and be like, or steal something.
197
00:15:53.367 --> 00:15:55.108
And so my mother was a writer.
198
00:15:55.208 --> 00:15:56.789
Her books were in the Books Three corpus.
199
00:15:56.809 --> 00:15:57.969
She wrote 19 books.
200
00:15:58.470 --> 00:15:59.150
They were in there.
201
00:15:59.430 --> 00:16:04.694
I don't think that anybody's, you know, necessarily prompting for romantic books of the 80s.
202
00:16:05.334 --> 00:16:06.095
necessarily.
203
00:16:06.395 --> 00:16:12.940
But if somebody wanted to, they could say, I want Palm Beach, but I want it 30% different or 10% different.
204
00:16:13.280 --> 00:16:29.392
And I think that that being able to dial in and reliably expect, you know, the amount of difference you're dialing in, if you're still using that source as the point of origin for your prompt, doesn't that feel, it feels like theft?
205
00:16:30.265 --> 00:16:31.386
Yeah, no, it absolutely is.
206
00:16:31.446 --> 00:16:39.690
And so let me take a step back, too, because you kind of combined two separate issues in terms of how the courts will consider this, right?
207
00:16:39.910 --> 00:16:50.716
This is exactly why I want to have this conversation, because I think that every time people say something like me or anybody else, they don't realize they're actually taking two completely disparate ideas and shoving them together.
208
00:16:50.756 --> 00:16:52.597
And so, yes, thank you.
209
00:16:52.757 --> 00:16:53.717
This is exactly what.
210
00:16:53.737 --> 00:16:57.719
They're not completely disparate, but the courts will treat them separately, or at least they should.
211
00:16:57.739 --> 00:16:58.600
And we'll see what happens.
212
00:16:59.200 --> 00:17:06.082
And so you mentioned, you said, well, wait a minute, it seems wrong that they're just copying your mother's book, right?
213
00:17:06.362 --> 00:17:09.163
And to produce something that might be 10% different.
214
00:17:09.403 --> 00:17:16.026
And so what the courts should look at is separate that into what I'll call input infringement and output infringement.
215
00:17:16.046 --> 00:17:17.126
They're two separate things.
216
00:17:17.826 --> 00:17:23.328
Under copyright law, when the AI companies copy your mother's book, they're not going
217
00:17:23.428 --> 00:17:26.749
That's a complete copy, period, end of story.
218
00:17:26.789 --> 00:17:28.569
That is copyright infringement.
219
00:17:28.730 --> 00:17:29.010
Right.
220
00:17:29.050 --> 00:17:36.092
The only way that's not copyright infringement is if they have a defense, like fair use, and why all these cases involve fair use.
221
00:17:36.492 --> 00:17:51.156
But you do not have to look at the actual AI output and compare the AI output to your mother's work to see is there an infringement or not, because there is definitely an infringement on the input side.
222
00:17:51.916 --> 00:17:58.100
What you would look at is does fair use would consider is the purpose of those two works the same?
223
00:17:58.200 --> 00:17:59.842
Are you producing a romance novel?
224
00:17:59.862 --> 00:17:59.922
No.
225
00:18:00.800 --> 00:18:03.021
Is your mother's work a romance novel?
226
00:18:03.061 --> 00:18:03.321
Yes.
227
00:18:03.642 --> 00:18:04.902
Is the output a romance novel?
228
00:18:04.922 --> 00:18:09.385
Well, the purpose is pretty much the same then, and so it's probably going to be unlikely to be fair use.
229
00:18:09.605 --> 00:18:11.046
Now, let me take the other infringement.
230
00:18:11.066 --> 00:18:22.792
The other infringement is output infringement, and that is when you do look at, is the output substantially similar to the inputted work, the ingested work?
231
00:18:23.213 --> 00:18:26.214
And that gets to your point if you change it by 10% or 20% or 50%.
232
00:18:27.955 --> 00:18:30.616
Who knows what their percentage is that it needs to be different?
233
00:18:30.776 --> 00:18:30.996
Yeah.
234
00:18:31.016 --> 00:18:32.777
But there are two separate infringements.
235
00:18:33.357 --> 00:18:36.198
So let's say you get something that's identical to your mother's book.
236
00:18:36.518 --> 00:18:38.759
There are two infringements there, right?
237
00:18:38.779 --> 00:18:41.920
There's infringement on the input side and on the output side.
238
00:18:42.180 --> 00:18:44.641
And a lot of people out there, I think, don't really understand that.
239
00:18:44.661 --> 00:18:51.423
They always think, oh, but the output is different, is different than what the AI companies are ingesting.
240
00:18:52.003 --> 00:18:53.144
And OK, that's helpful.
241
00:18:53.485 --> 00:18:57.728
Oh, but it's helpful when you're talking about output infringement, not the input infringement.
242
00:18:58.109 --> 00:18:59.450
Hopefully that makes some amount of sense.
243
00:19:00.110 --> 00:19:00.971
Yes, absolutely.
244
00:19:00.991 --> 00:19:11.200
So are you saying that that by simply digitizing, just taking a book, digitizing it, you've already done something illegal or morally dubious?
245
00:19:11.941 --> 00:19:12.422
Absolutely.
246
00:19:12.442 --> 00:19:16.065
That initial that act itself is copyright infringement.
247
00:19:16.413 --> 00:19:25.918
Yeah, and you don't have to look any further than I think it was a year and a half ago, maybe two years ago now, where the book publishers brought a case against a company organization called the Internet Archives.
248
00:19:26.478 --> 00:19:28.299
And in that case, that was in the Second Circuit.
249
00:19:28.319 --> 00:19:33.722
That's the case I was referring to before, where the court said, you're digitizing these books.
250
00:19:34.122 --> 00:19:35.583
You don't have authority to do that.
251
00:19:35.623 --> 00:19:36.783
That's not fair use.
252
00:19:37.264 --> 00:19:39.385
It's an infringement, period, end of story.
253
00:19:39.405 --> 00:19:43.427
So yes, the mere fact of digitizing these works without permission is,
254
00:19:44.267 --> 00:19:47.288
without compensation will be considered to be an infringement.
255
00:19:47.809 --> 00:19:54.611
In certain instances, maybe it might be considered to be a fair use or certainly cases out there where it has been held to be a fair use.
256
00:19:55.592 --> 00:20:03.275
But it will be an infringement and the company that does it will be liable unless the fair use defense would apply.
257
00:20:03.515 --> 00:20:04.836
That's fascinating.
258
00:20:04.896 --> 00:20:16.184
And that's something I absolutely didn't know, because in my mind, and especially because we sort of see clips of things on YouTube and whatever else, it's always been in my mind that you can sort of do whatever you want behind the scenes.
259
00:20:16.224 --> 00:20:21.087
The only time it becomes official copyright infringement is if you monetize something.
260
00:20:21.567 --> 00:20:46.126
that you've you've tried to sell something that isn't yours and that actually by clipping your favorite trailers of the year on putting it on youtube on a non-monetized thing that actually you've sort of just been a part of you know the conversation of culture you haven't actually benefited or profited from it and so you haven't committed any sort of copyright theft but but it sounds like that's
261
00:20:47.413 --> 00:20:53.356
a myth in the either I've created or colloquially we've all created as a as a populace.
262
00:20:53.836 --> 00:20:54.956
Yeah, no, I think that's right.
263
00:20:54.976 --> 00:21:00.058
What you what you're saying, because copyright law, and it doesn't require any state of mind, right?
264
00:21:00.078 --> 00:21:03.320
You don't have to know you're engaging in infringement or anything like that.
265
00:21:03.760 --> 00:21:06.122
You don't have to be charging money for it.
266
00:21:06.142 --> 00:21:08.003
You don't have to be profiting or anything.
267
00:21:08.163 --> 00:21:15.088
The mere fact that you are copying something, that is a violation of a copyright owner's reproduction right.
268
00:21:15.609 --> 00:21:22.113
And that person who does the copying, who engages in the copying, will be liable unless there's a defense.
269
00:21:22.374 --> 00:21:26.437
Now, maybe there's a defense of some sort that applies in the AI case.
270
00:21:26.817 --> 00:21:32.321
That defense is at issue in about 120 cases is the defense of fair use.
271
00:21:32.681 --> 00:21:33.382
OK, then.
272
00:21:33.402 --> 00:21:36.545
Well, that I mean, listen, that's I've learned something every day is a school day.
273
00:21:36.665 --> 00:21:45.853
I've learned something absolutely major and I'll be very careful in the future how I dress because I've I've made that mistake so many times on this podcast.
274
00:21:46.253 --> 00:21:49.876
So I'm thrilled to have learned something new regarding output.
275
00:21:49.936 --> 00:21:53.699
And I love that you've made this distinction between input and output, but regarding output.
276
00:21:54.764 --> 00:21:56.446
How much does the prompting matter?
277
00:21:56.487 --> 00:22:08.183
We've heard from some people that you need to be logging all of your prompts so that it's clear that you say, oh, I want a muted color scheme and I want a really balanced framing.
278
00:22:08.203 --> 00:22:09.224
And you can say that.
279
00:22:09.304 --> 00:22:11.828
What you can't say is I want it to look like Wes Anderson.
280
00:22:12.148 --> 00:22:12.388
Okay.
281
00:22:12.688 --> 00:22:12.828
Yep.
282
00:22:13.108 --> 00:22:25.693
When it comes to defending the output or claiming copyright on something that you've been a part of creating using AI, what are the levels of intervention at play?
283
00:22:25.933 --> 00:22:31.195
What does one have to do to be protecting or using these sorts of tools?
284
00:22:31.556 --> 00:22:34.157
To be clear, we're going to switch topics now, right?
285
00:22:34.177 --> 00:22:35.618
We have been talking about infringement.
286
00:22:36.058 --> 00:22:38.820
I understand you're now asking about copyrightability, right?
287
00:22:38.860 --> 00:22:41.921
If you're using these tools, is the end result?
288
00:22:42.021 --> 00:22:44.863
Is the output protected by copyright, right?
289
00:22:44.903 --> 00:22:45.463
Well, sure.
290
00:22:45.523 --> 00:22:52.387
Actually, in that case, should we take the first step, which would be what point is it infringement along that output path?
291
00:22:52.867 --> 00:22:55.448
Yeah, I mean, and the test would be substantial similarity.
292
00:22:55.468 --> 00:22:55.769
Sure.
293
00:22:55.789 --> 00:22:56.489
Is the output protected?
294
00:22:56.829 --> 00:23:01.892
substantially similar to the input, if you will, to a copyrighted work.
295
00:23:02.172 --> 00:23:03.213
And that's the test.
296
00:23:03.273 --> 00:23:09.776
It's been the test for a very, very long time in terms of determining what is and what is not infringing.
297
00:23:09.976 --> 00:23:19.742
And so if you're talking about copyrightability, then what you do is you look at, and from the copyright office's standpoint and the court so far,
298
00:23:20.898 --> 00:23:23.761
They basically say, look, there's got to be a human.
299
00:23:24.722 --> 00:23:26.123
An author's got to be a human.
300
00:23:26.163 --> 00:23:27.405
It cannot be AI, right?
301
00:23:27.465 --> 00:23:33.631
So if you ask AI to do something simple like, you know, make me a cartoon frog or something like that.
302
00:23:33.651 --> 00:23:35.993
Well, that's not going to be protected.
303
00:23:36.113 --> 00:23:36.474
Period.
304
00:23:36.534 --> 00:23:37.114
End of story.
305
00:23:37.955 --> 00:23:53.146
If you put in pages and pages of prompts and what have you, the viewpoint still is that that end result is not going to be protected because it's AI that made the decisions in terms of how to interpret your prompts.
306
00:23:53.566 --> 00:24:01.391
That could change over time, but that's what the law is right now in terms of because it's going to be not protected by copyright.
307
00:24:01.712 --> 00:24:05.815
But then you have the mix of let's say you're writing a book
308
00:24:07.035 --> 00:24:13.457
and you wrote your book, and you say, well, the characters' names aren't terribly interesting here.
309
00:24:13.497 --> 00:24:21.078
I'd like to – let me put this through AI and say, AI, give me some more interesting characters or change – or something like that.
310
00:24:21.818 --> 00:24:26.199
Well, you did most of the work there, and all it did was add and change some names.
311
00:24:26.459 --> 00:24:28.540
So that is going to be protected by copyrights.
312
00:24:29.060 --> 00:24:31.181
But you can imagine you've got a scale here.
313
00:24:31.201 --> 00:24:40.327
You've got this big gray area in between where we just don't know if you've done enough to actually warrant copyright protection.
314
00:24:40.387 --> 00:24:40.587
Right.
315
00:24:40.607 --> 00:24:42.168
And that is very much up for debate.
316
00:24:42.208 --> 00:24:53.134
There's one case that was just decided, but that was an easy case where the person who was the so-called creator said, I did nothing.
317
00:24:53.334 --> 00:24:54.255
A, I did it all.
318
00:24:54.735 --> 00:24:55.716
Well, that's an easy quiz.
319
00:24:55.736 --> 00:24:57.677
That's the green frog case I mentioned before.
320
00:24:57.697 --> 00:24:57.797
Yeah.
321
00:24:57.817 --> 00:24:57.997
Yeah.
322
00:24:58.257 --> 00:25:05.364
But there's a case pending right now in the courts where the person did a little bit of the artwork themselves and AI did a little bit of it.
323
00:25:05.424 --> 00:25:11.530
And it becomes a lot more difficult to figure out, okay, what's protected and what is not protected there.
324
00:25:12.131 --> 00:25:23.122
Is there a certain point at which the complexity of the prompt, let's say, and we'll just use a very simple example, an absolutely massive complex prompt and one single output.
325
00:25:23.662 --> 00:25:27.506
Is there a level at which the complexity of the prompt becomes protectable?
326
00:25:27.786 --> 00:25:31.590
Because obviously, if I do a make me a cartoon frog, that's another thing.
327
00:25:31.850 --> 00:25:33.512
If I write the wind in the willows.
328
00:25:35.093 --> 00:25:42.438
And say this is taking place in a fictional world and there is a frog who's the leader of it and whatever else.
329
00:25:43.139 --> 00:25:46.962
Well, that's now what could be argued as a novel or a novel series.
330
00:25:47.342 --> 00:25:50.545
Is only my protection of those words protected?
331
00:25:51.125 --> 00:26:00.132
Once it becomes an AI output, it's sullied by the training data or the decisions of the model and it no longer becomes my protectable work.
332
00:26:00.212 --> 00:26:00.392
Right.
333
00:26:00.552 --> 00:26:02.433
So it would still be protected, right?
334
00:26:02.473 --> 00:26:09.775
Your work would be protected, but anything that was added, that was generated by AI would not be protected, right?
335
00:26:09.915 --> 00:26:10.135
Right.
336
00:26:10.155 --> 00:26:10.935
So let's take my green frog.
337
00:26:10.955 --> 00:26:11.695
So anything visual.
338
00:26:11.975 --> 00:26:14.056
Well, so let's take my green frog example, right?
339
00:26:14.356 --> 00:26:17.197
So let's say I drew a green frog and...
340
00:26:18.537 --> 00:26:19.917
you know, an animated green frog.
341
00:26:20.257 --> 00:26:25.699
And then I put it into the AI system and said, AI, can you put clothes on my frog and put a hat on him or something?
342
00:26:26.179 --> 00:26:30.820
That part's not going to be protected, but my green frog that I drew myself will be protected.
343
00:26:31.460 --> 00:26:41.543
So it's going to be a mix of, they're going to be, it's going to be a copyrighted work at the end of the day, but there are elements of that copyrighted work that are not yours, that are not going to be protectable.
344
00:26:41.923 --> 00:26:44.084
And that's not crazy unique.
345
00:26:44.124 --> 00:26:46.446
I mean, take a motion picture, for example.
346
00:26:46.766 --> 00:26:48.487
You go to see a movie in the theaters.
347
00:26:48.507 --> 00:26:51.769
There's music playing within the movie.
348
00:26:52.229 --> 00:26:55.091
And that music, let's say, was created by some other artist.
349
00:26:55.471 --> 00:26:59.473
Well, the motion picture studio is not going to claim copyright in that song.
350
00:27:00.114 --> 00:27:01.975
That belongs to someone else.
351
00:27:02.255 --> 00:27:07.118
So this idea of separating out the elements that you did not create
352
00:27:07.738 --> 00:27:13.099
and someone else has created it or something, in this case with AI, that's not a unique aspect of copyright law.
353
00:27:13.119 --> 00:27:14.740
That's been done before.
354
00:27:14.780 --> 00:27:16.340
It happens all the time.
355
00:27:17.380 --> 00:27:20.841
It just is more convoluted and complex, I think, with AI.
356
00:27:21.021 --> 00:27:31.523
Yeah, there was an interesting case that I got to sort of see up close where the rights to Terminator were going to revert back to James Cameron after 50 years or something like that.
357
00:27:32.183 --> 00:27:34.304
And what was interesting was...
358
00:27:34.944 --> 00:27:37.146
what the rates were that were reverting back.
359
00:27:37.647 --> 00:27:43.613
Any characters that were part of his original script and his Terminator 2 script reverted back.
360
00:27:43.933 --> 00:27:52.402
But any characters that were created from Terminator 3 onwards were not covered by that because they were not, those characters were not within.
361
00:27:52.442 --> 00:27:54.665
Now, arguably Cameron goes, I don't care.
362
00:27:54.725 --> 00:27:57.548
None of the characters beyond Terminator 2 were any good.
363
00:27:57.988 --> 00:28:08.996
But it was very interesting seeing that delineation in real time in something as traditional as film and screenwriting that, you know, certain things are coming back, certain things are not.
364
00:28:09.376 --> 00:28:19.563
And I think in the first tranche of rights that went back to him, it was only that original script, that original movie, Sarah Connor, the idea of John Connor, but not the T-1000, only the T-800.
365
00:28:19.603 --> 00:28:23.546
Sorry, like I said, with all these tchotchkes behind me, I'm quite a nerd.
366
00:28:23.566 --> 00:28:23.706
Yeah.
367
00:28:24.767 --> 00:28:25.007
Yeah.
368
00:28:25.428 --> 00:28:27.610
But but I followed that quite avidly.
369
00:28:27.650 --> 00:28:33.876
And it's a good example, I think, of what you're talking about, which is you have certain rights to certain things.
370
00:28:33.896 --> 00:28:40.742
But depending on when and how they're manifested, you may not own the characters surrounding your primary idea.
371
00:28:40.803 --> 00:28:42.945
You know, the the family of frogs.
372
00:28:43.585 --> 00:28:47.990
If the AI comes up with a family of frogs, you create this great storybook.
373
00:28:48.935 --> 00:28:52.338
world and then actually you only own the primary frog.
374
00:28:52.699 --> 00:28:52.939
Yeah.
375
00:28:52.959 --> 00:28:53.099
Yeah.
376
00:28:53.159 --> 00:28:55.561
No, I think that's, I think that's generally right.
377
00:28:55.942 --> 00:28:58.504
And, and that's why these issues get a little complicated games.
378
00:28:58.524 --> 00:29:04.790
Cameron examples, probably frankly, all done under contract contractual terms and, and, and what have you.
379
00:29:04.850 --> 00:29:07.793
And so that becomes even a little bit more complex, but yeah,
380
00:29:07.893 --> 00:29:12.956
But certainly, yeah, I mean, that's why these issues are issues of first impression.
381
00:29:12.976 --> 00:29:14.076
People will ask questions.
382
00:29:14.096 --> 00:29:15.317
What about this and what about that?
383
00:29:15.917 --> 00:29:20.079
And me and my team and other attorneys give the best answer we can.
384
00:29:20.539 --> 00:29:23.201
But we also realize that that could change on a dime.
385
00:29:23.281 --> 00:29:26.682
Tomorrow we could get some court case that says, here's the standard.
386
00:29:26.882 --> 00:29:28.243
Here's what you need to be looking for.
387
00:29:28.403 --> 00:29:31.185
Or maybe Congress passes some law that does the same thing.
388
00:29:31.525 --> 00:29:33.806
But right now, today, people.
389
00:29:34.166 --> 00:29:43.428
Yeah, that's I mean, we have to look at each individual case and understand the facts in those cases and how the copyright law would apply to the facts in that case.
390
00:29:43.868 --> 00:29:49.709
Yeah, you said something on our briefing call that I felt was very important, which is AI is not going away.
391
00:29:49.749 --> 00:29:53.169
So the question is, how do we solve the copyright issues around it?
392
00:29:53.489 --> 00:29:56.370
What are the main things that a solution actually has to solve?
393
00:29:56.790 --> 00:29:58.330
So I think there's three things, right?
394
00:29:58.410 --> 00:30:03.211
And there's one that's one of the three licensing is the most important issue.
395
00:30:03.711 --> 00:30:15.355
I think if you have free market licensing where AI companies get licenses from copyright owners to ingest their works and train on their works, I think everyone goes home happy, basically, right?
396
00:30:15.715 --> 00:30:16.996
So licensing is number one.
397
00:30:17.276 --> 00:30:23.338
You need some element of transparency so copyright owners can know if their works are being copied or not.
398
00:30:23.438 --> 00:30:25.239
So you need some element of transparency.
399
00:30:25.599 --> 00:30:26.979
I don't think you get that in the courts.
400
00:30:27.019 --> 00:30:28.440
That will have to come through Congress.
401
00:30:29.580 --> 00:30:32.581
And then last thing is what I'll call safeguards, right?
402
00:30:32.881 --> 00:30:46.325
You need each AI company needs to implement safeguards to make sure that works or whatever's output by the AI system is not identical or substantially similar to a copyrighted work.
403
00:30:46.345 --> 00:30:51.926
In other words, you're not producing Darth Vader or SpongeBob SquarePants or whatever the case may be.
404
00:30:52.286 --> 00:30:53.427
And there needs to be safeguards.
405
00:30:53.467 --> 00:30:56.007
And some companies are doing that and some are not.
406
00:30:58.788 --> 00:31:04.530
I think probably the big three elements, I mean, licensing, safeguards, and transparency.
407
00:31:05.150 --> 00:31:10.592
What you just said is very interesting because copyright has always had to serve two masters, hasn't it?
408
00:31:11.073 --> 00:31:21.456
On the one hand, it has to protect artists and creators enough that they can make a living, but it also needs to not lock culture so tightly that new creativity becomes impossible.
409
00:31:21.897 --> 00:31:24.478
And so it's sort of a little bit like the U.S. Constitution.
410
00:31:24.778 --> 00:31:25.178
It should be...
411
00:31:25.938 --> 00:31:30.582
Rigid enough to give us guidance, but stretchy enough that it allows for growth.
412
00:31:31.123 --> 00:31:41.172
And I think, you know, again, I'm sorry if I bring up nerdy examples, but, you know, the original Teenage Mutant Ninja Turtles was essentially a ripoff of Daredevil.
413
00:31:41.492 --> 00:31:45.373
You know, in Daredevil, you had the hand that was the all the bad guys.
414
00:31:45.393 --> 00:31:48.314
And in Teenage Mutant Ninja Turtles, you had the Foot Clan.
415
00:31:48.694 --> 00:31:53.335
And it was, you know, there was all this sort of playful stealing or borrowing.
416
00:31:53.515 --> 00:31:54.676
And that's a good thing.
417
00:31:54.716 --> 00:31:57.316
And copyright is built to protect that.
418
00:31:57.877 --> 00:32:02.678
Is there a rebalancing of those two needs now?
419
00:32:02.838 --> 00:32:04.439
around the AI conversation?
420
00:32:04.679 --> 00:32:06.059
Do we need to be more strict?
421
00:32:06.300 --> 00:32:14.903
Yeah, I think we're already seeing that happen a little bit, is that there was a Supreme Court case decided a few years ago in the Andy Warhol case.
422
00:32:15.904 --> 00:32:18.745
And what the Supreme Court said is, wait a minute,
423
00:32:19.245 --> 00:32:20.867
Fair use has kind of gotten too far.
424
00:32:20.967 --> 00:32:22.449
It's gotten too large.
425
00:32:22.469 --> 00:32:27.955
This idea of transformative use has swallowed up the entire fair use analysis, right?
426
00:32:28.736 --> 00:32:31.519
And we need to rein that in a little bit, right?
427
00:32:31.740 --> 00:32:33.241
And that's what the court did in that case.
428
00:32:33.281 --> 00:32:36.225
And so we've already seen a little bit of the kind of reining in.
429
00:32:36.245 --> 00:32:39.429
And fair use is this four-pronged test.
430
00:32:39.849 --> 00:32:44.490
that is intended to balance the very things that you were talking about.
431
00:32:44.550 --> 00:32:51.651
On the one hand, you want to incentivize creativity and incentivize people to distribute their copyrighted works far and wide.
432
00:32:52.511 --> 00:32:55.252
And the way to do that, frankly, is they get paid for it then, right?
433
00:32:56.112 --> 00:33:07.494
And so you want to create that incentive alive at the same, by the same token, you want to incentivize other people to look at those works and see those works and to inspire them and for them to create new works from that.
434
00:33:07.934 --> 00:33:09.775
And that is a difficult balance.
435
00:33:09.796 --> 00:33:11.577
And we talked about pendulums before.
436
00:33:11.877 --> 00:33:13.718
Fair use is most definitely a pendulum.
437
00:33:13.758 --> 00:33:19.363
Sometimes it goes a little too far on one side or the other, and then it kind of swings back to the other side.
438
00:33:19.623 --> 00:33:28.950
That's exactly what was happening in the Andy Warhol Foundation case, where the pendulum was, the court says, wait a minute, this pendulum swung a little bit too far to the fair use side.
439
00:33:29.230 --> 00:33:34.294
We need to bring this back a little bit more to protect creators to keep this incentive going.
440
00:33:34.914 --> 00:33:38.896
And that's ultimately, I think, what the big takeaway from that case is.
441
00:33:39.496 --> 00:33:48.579
And I think we're seeing that in the AI context now as well, trying to figure out how fair use should apply in the AI context.
442
00:33:48.719 --> 00:33:50.600
And, you know, we at the Copper Alliance are...
443
00:33:51.220 --> 00:33:53.101
are big supporters of AI, right?
444
00:33:53.141 --> 00:33:59.404
As long as it's done responsibly, ethically, and respectfully, right?
445
00:34:00.204 --> 00:34:05.727
On the one hand, but for fair use, we understand that some of these cases probably will be fair use.
446
00:34:05.907 --> 00:34:09.829
They will be legitimate fair use, but the vast majority of them probably will not be.
447
00:34:09.849 --> 00:34:12.530
A lot of the cases that we're seeing in the courts today
448
00:34:13.150 --> 00:34:18.152
I think the vast majority of them, the court should come out and say, no, no, this is not fair use.
449
00:34:18.172 --> 00:34:24.654
You are copying a work in order to compete with and frankly destroy the market for the work that you're copying.
450
00:34:24.714 --> 00:34:27.334
And you're doing it without permission and without compensation.
451
00:34:27.714 --> 00:34:28.835
And that just shouldn't be allowed.
452
00:34:28.855 --> 00:34:32.796
You're going to destroy whole areas of creativity.
453
00:34:33.096 --> 00:34:35.437
And where do you guys fit in that conversation?
454
00:34:35.557 --> 00:34:39.738
How do you, you know, you said your mission is to protect artists, creatives and creators from,
455
00:34:39.958 --> 00:34:45.441
Where do you find you're doing the most good and getting the most impact in that mission?
456
00:34:45.701 --> 00:34:51.143
I mean, so part of it is where we're having, I think, I hope, the most impact is, for instance, what we're doing today.
457
00:34:51.544 --> 00:34:58.627
It's just explaining to people copyright law and explaining how copyright law applies to AI and to other things.
458
00:34:59.247 --> 00:35:02.209
And so there's certainly an element of education.
459
00:35:02.509 --> 00:35:08.692
There's also an element of advocacy, going up on Capitol Hill, going to the executive branch and explaining that.
460
00:35:08.932 --> 00:35:14.297
how fair use applies, what the purpose of fair use is, what the purpose of copyright law is.
461
00:35:14.737 --> 00:35:17.860
Do we need new laws here, for instance, with Congress?
462
00:35:18.080 --> 00:35:20.602
And we do the same thing with the courts.
463
00:35:20.682 --> 00:35:25.166
We'll file amicus brief with the courts explaining this is how the law has applied.
464
00:35:25.206 --> 00:35:30.150
In other cases, maybe this is right, this is wrong, things of that nature.
465
00:35:30.190 --> 00:35:34.214
So part of his advocacy, part of it is education.
466
00:35:34.574 --> 00:35:35.255
And I also –
467
00:35:35.895 --> 00:35:38.458
Should add that we also educate our members, right?
468
00:35:38.478 --> 00:35:47.869
When AI first came along, a lot of our members, just especially our smaller members, like the individual photographers and what have you, didn't necessarily understand AI and how it worked.
469
00:35:47.909 --> 00:35:53.436
And so part of our job is to educate them so they can protect themselves as best as possible.
470
00:35:54.557 --> 00:35:56.560
What are creatives experiencing, right?
471
00:35:56.641 --> 00:35:57.602
Artists, creatives.
472
00:35:57.682 --> 00:36:09.560
I mean, because on the one hand, I saw a stat the other day and it was mainly focused in England, but it was that the creative industries are growing at nearly two times the rest of the British economy.
473
00:36:10.101 --> 00:36:19.869
But what was interesting is I was there and I was meeting with lots of people who ran production companies and studios and advertising agencies, and they are not feeling that growth.
474
00:36:20.410 --> 00:36:35.523
And so it seems that so much of the growth has gone from creatives or creatives that work with inside industry or artists who people are creatives that are sort of on their own to create tours who are, you know, kind of creative.
475
00:36:35.663 --> 00:36:40.973
taking advantage of direct distribution to kind of, you know, make money online and in socials.
476
00:36:41.354 --> 00:36:45.402
We're definitely seeing that pain for the artists and the creatives.
477
00:36:45.422 --> 00:36:46.083
Yeah.
478
00:36:46.295 --> 00:37:01.465
You know, what are you seeing in terms of their experience now in the first major shift, which is the disruption of distribution versus the new shift, which is the disruption of production, which is what we're seeing with AI.
479
00:37:03.166 --> 00:37:04.407
I don't know what's happening today.
480
00:37:04.427 --> 00:37:06.608
Sorry, Keith.
481
00:37:06.668 --> 00:37:07.869
I'm tripping over my own mouth.
482
00:37:08.249 --> 00:37:10.931
This is what happens when I ask seven part questions.
483
00:37:10.951 --> 00:37:11.791
So it's entirely my fault.
484
00:37:11.811 --> 00:37:12.312
No, I mean, look.
485
00:37:12.652 --> 00:37:13.933
What are creatives experiencing?
486
00:37:13.953 --> 00:37:14.213
I don't know.
487
00:37:14.673 --> 00:37:19.757
I wish I could put all creatives or creators in a bucket and say, this is what they're experiencing.
488
00:37:20.077 --> 00:37:21.918
This is what their belief system is.
489
00:37:22.659 --> 00:37:23.979
But that's just not possible.
490
00:37:24.860 --> 00:37:33.105
If you survey all of the close to 2 million individual creators that we represent, you're going to get a variety of views.
491
00:37:33.186 --> 00:37:36.027
Some of which you'll say, you know, AI is terrible.
492
00:37:36.087 --> 00:37:37.228
I would never use AI.
493
00:37:37.328 --> 00:37:41.471
Others that absolutely do use AI for things like ideation.
494
00:37:41.531 --> 00:37:42.372
I gave the example.
495
00:37:43.172 --> 00:37:50.079
of someone who wrote like a short story or a book and wanted more interesting names for their characters.
496
00:37:50.720 --> 00:37:53.623
That comes from experience, from talking to a creator who did that, right?
497
00:37:54.243 --> 00:37:54.464
Yeah.
498
00:37:54.644 --> 00:37:56.245
And so there's a little bit of a mix.
499
00:37:56.325 --> 00:38:03.553
And to be honest, you go onto some of these forums where creators are talking and they may go at each other a little bit, you know, in terms of,
500
00:38:04.013 --> 00:38:12.515
well, you're disenfranchising us because you're using AI to help with your creativity and what have you.
501
00:38:12.755 --> 00:38:16.195
And so they'll get into arguments themselves.
502
00:38:16.215 --> 00:38:22.917
So I wish I could put all creatives or creators into a bucket and say this is how they feel, but that just really isn't the case.
503
00:38:22.937 --> 00:38:24.337
You're getting a wide variety of views.
504
00:38:24.697 --> 00:38:28.418
I think the one consistency that you hopefully see is –
505
00:38:29.198 --> 00:38:37.924
but this is kind of hard to implement in practice is creators do want to be respectful and they want to use AI that is respectful of copyright.
506
00:38:37.944 --> 00:38:43.168
Like there are lots of AI companies out there that are licensing.
507
00:38:43.248 --> 00:38:52.955
We have a page on our website that goes through thousands of licenses between AI companies and copyright owners that are, that are being agreed to.
508
00:38:53.195 --> 00:39:05.446
And so I think that small individual creators do try their hardest to look, okay, is this a legitimate AI company that's trying to do the right thing, trying to license?
509
00:39:06.267 --> 00:39:09.530
Or is it just stealing my stuff and everyone else's stuff?
510
00:39:09.850 --> 00:39:12.312
So I think they do try to use the...
511
00:39:13.333 --> 00:39:14.754
technologies that are respectful.
512
00:39:14.794 --> 00:39:21.197
And there's a lot of, and this doesn't get said very often, but there's a lot of smaller AI companies like Bria AI is a good example.
513
00:39:22.198 --> 00:39:23.779
Pro Rata is another example.
514
00:39:23.819 --> 00:39:33.644
There's a few others that are licensing works from copyright owners and creators in order to train their AI systems.
515
00:39:34.044 --> 00:39:37.946
It's really the big guys that can afford all these, to hire all these attorneys.
516
00:39:38.286 --> 00:39:41.448
You know, it's the metas of the world and the Googles of the world and, you know,
517
00:39:41.688 --> 00:40:08.507
uh anthropics and so on and so forth that have billions upon billions of dollars to throw at litigation uh that they're willing they're willing to do that rather than license for the most part but but a lot of other ai companies are licensing who are you seeing so you know like i appreciate you can't lump everybody in together but who are you seeing that you feel is particularly at threat because if it were me i would be looking at this the creative community and the artistic community and i'd be like i wouldn't want to be a writer right now
518
00:40:10.208 --> 00:40:14.770
You know, for example, the Mia Ballard case where, you know, she wrote a book.
519
00:40:14.870 --> 00:40:15.931
It did very well.
520
00:40:16.371 --> 00:40:22.874
And then she was accused of using AI and her publisher just pulled the book and sort of, you know, left her on her own.
521
00:40:22.894 --> 00:40:23.735
And I thought that was...
522
00:40:24.635 --> 00:40:32.679
Shocking and unfair to somebody who, regardless of how she came out with it, the publisher, I think, should have backed her through that process.
523
00:40:32.699 --> 00:40:37.442
I think abandoning her was incredibly, you know, rude and inappropriate thing to do.
524
00:40:37.562 --> 00:40:40.443
You know, they they either failed in their own vetting practice.
525
00:40:41.124 --> 00:40:46.469
or just succumb to allowing her to kind of be execution style canceled.
526
00:40:46.890 --> 00:40:53.296
But I, you know, I think it speaks to this thing that writers do a very hard thing that has now become substantially easier.
527
00:40:53.336 --> 00:40:58.222
And I don't think that AI should solve writing, but I do think it can solve writer's block.
528
00:40:58.762 --> 00:40:59.303
Yeah.
529
00:40:59.323 --> 00:41:04.290
And but yeah, it's sort of changing the dynamics of what's what what is valuable.
530
00:41:05.692 --> 00:41:06.293
Yeah.
531
00:41:06.994 --> 00:41:08.075
That was just my assessment.
532
00:41:08.095 --> 00:41:09.417
Sorry, I answered my own question.
533
00:41:09.437 --> 00:41:14.785
But are there specific creators or artists that you think are particularly at risk?
534
00:41:15.045 --> 00:41:15.866
I think there are.
535
00:41:17.146 --> 00:41:22.050
I think just to pull one example, because we've talked about it already, is romance writers.
536
00:41:22.130 --> 00:41:25.252
I think romance writers, I have to admit, I haven't read any romance books.
537
00:41:25.272 --> 00:41:26.592
Because of my mom's books being in there.
538
00:41:26.733 --> 00:41:28.234
Yeah, yeah, yeah.
539
00:41:28.594 --> 00:41:31.476
But from my understanding, it's fairly formulaic.
540
00:41:31.616 --> 00:41:35.978
I apologize to all the romance writers I've not read, if that's not the case.
541
00:41:36.259 --> 00:41:39.781
But from what I've heard from other people, that's, I think, a good example of...
542
00:41:40.101 --> 00:41:40.441
Yeah.
543
00:41:40.562 --> 00:41:41.422
Computer coders.
544
00:41:41.522 --> 00:41:42.303
Right.
545
00:41:42.443 --> 00:41:43.844
Is another example of that.
546
00:41:43.864 --> 00:41:46.947
I've heard quite a bit that they are they are a threat.
547
00:41:46.987 --> 00:41:49.630
And then there's other examples that that's such a good point.
548
00:41:49.670 --> 00:41:52.252
We think about copyright and we naturally think of artists.
549
00:41:52.272 --> 00:41:53.533
But you're absolutely right.
550
00:41:53.553 --> 00:41:55.275
I mean, computer code is copyrightable.
551
00:41:55.755 --> 00:41:58.538
Yeah, yeah, no, absolutely, absolutely.
552
00:41:58.578 --> 00:42:08.948
So, I mean, and that brings me to one thing I want to mention, which is, you know, when we talk about AI, we talk about it in terms of how it, or generative AI, I should say.
553
00:42:09.189 --> 00:42:17.117
When we talk about generative AI, we talk about it as if it applies equally across the board, whether you're talking about a movie or a book or a book.
554
00:42:17.777 --> 00:42:19.478
or, or music.
555
00:42:19.858 --> 00:42:21.878
And that is just not the case, right?
556
00:42:21.958 --> 00:42:26.840
Is that it is going to apply differently depending on the type of work at issue.
557
00:42:26.880 --> 00:42:27.840
And let me give an example.
558
00:42:28.481 --> 00:42:28.781
Okay.
559
00:42:28.821 --> 00:42:33.022
You could be, for instance, copying, and I'll just stick with the romance novels here.
560
00:42:33.262 --> 00:42:35.403
You could be copying all the romance novels,
561
00:42:36.123 --> 00:42:46.034
In the world, okay, to create an AI system, but that AI system is used only to write cover letters or resumes or thank you notes or something.
562
00:42:46.554 --> 00:42:48.596
That's not going to be competitive.
563
00:42:49.237 --> 00:42:52.620
No one is going to be licensing the romance novels for that purpose.
564
00:42:53.521 --> 00:42:55.884
So that's very likely going to be a fair use.
565
00:42:56.264 --> 00:42:56.625
However...
566
00:42:57.567 --> 00:43:10.493
If you were copying all these romance novels to make more romance novels that compete with the authors of those romance novels who you copied, then that very likely isn't going to be a fair use because that's going to harm the market.
567
00:43:11.033 --> 00:43:17.656
And so you saw in text, it's more difficult analysis, I think, at least.
568
00:43:18.076 --> 00:43:19.056
But let's look at music.
569
00:43:19.116 --> 00:43:20.417
And music, for instance...
570
00:43:21.317 --> 00:43:23.980
You're copying music for what?
571
00:43:24.260 --> 00:43:25.361
To make more music.
572
00:43:25.541 --> 00:43:27.143
I mean, that's what you're doing, right?
573
00:43:27.283 --> 00:43:37.754
And that is inherently going to be harmful and competitive and serve the same purpose, if you will, as other music.
574
00:43:38.094 --> 00:43:41.477
So I think the music cases, the newspaper cases...
575
00:43:42.859 --> 00:43:47.462
are and some of the image cases are really, really strong cases.
576
00:43:47.522 --> 00:43:53.147
I would be shocked if any of the music companies, music plaintiffs lost their cases.
577
00:43:53.267 --> 00:43:54.228
And same with the newspapers.
578
00:43:54.768 --> 00:43:59.592
And they have very, very strong cases for the example I just gave.
579
00:44:00.012 --> 00:44:02.915
So what is it about it that makes their case so strong?
580
00:44:03.115 --> 00:44:10.381
And is there something that other creatives and creators and IP owners can strengthen their case by learning from that?
581
00:44:10.681 --> 00:44:10.961
Yeah.
582
00:44:11.421 --> 00:44:19.505
So let's go back to something I said before about the input and the fact that AI companies are copying entire work.
583
00:44:19.525 --> 00:44:21.846
So it's an infringement unless there's a defense.
584
00:44:22.186 --> 00:44:23.267
Defense could be fair use.
585
00:44:23.787 --> 00:44:28.029
So basically, it's all is going to come down to is it fair use or is it not fair use?
586
00:44:28.734 --> 00:44:31.015
Now, if you're looking, let's take the music case.
587
00:44:31.095 --> 00:44:40.539
If you're looking at fair use, whether something is a fair use, a lot of times it depends on is there going to be market harm here and what is the purpose of the use.
588
00:44:40.659 --> 00:44:44.861
If the purpose is completely different, then there's a very good chance it's going to be a fair use.
589
00:44:45.321 --> 00:44:55.110
So in the example I gave with the romance novels, if you're making resumes, if you've got a tool to make resumes and cover letters, well, the purpose is very different than the purpose of a romance novel.
590
00:44:55.851 --> 00:45:00.515
If you're in the music side, if you're copying music to make other music,
591
00:45:01.415 --> 00:45:03.056
It's going to cause market harm.
592
00:45:03.076 --> 00:45:06.357
It's a very linear one-to-one relationship.
593
00:45:06.397 --> 00:45:06.617
Yeah.
594
00:45:06.817 --> 00:45:07.037
Yeah.
595
00:45:07.338 --> 00:45:15.841
And so that's why I think the music cases, newspaper cases are a similar boat, are very, very, very, very strong cases, and they should prevail in those cases.
596
00:45:16.061 --> 00:45:22.403
So it's sort of the diversification of output, you know, sort of protect some of the language models.
597
00:45:22.443 --> 00:45:26.064
It's like, well, no, no, no, but this thing is here to, you know, write a contract.
598
00:45:26.104 --> 00:45:27.325
It's here to schedule your day.
599
00:45:27.365 --> 00:45:28.525
It's here to be your coach.
600
00:45:28.605 --> 00:45:30.286
It's here to do all of these things.
601
00:45:30.806 --> 00:45:32.686
And yeah, it can help you write something.
602
00:45:32.726 --> 00:45:38.868
But, you know, so, you know, but the music is we ingested this for a single purpose.
603
00:45:39.048 --> 00:45:40.049
Yeah, yeah, yeah.
604
00:45:40.109 --> 00:45:41.189
And that's very interesting.
605
00:45:41.409 --> 00:45:50.394
And this is why I think, going back to my earlier point, too, about where people get confused about, oh, wait a minute, you have to look at the output, whether the output is substantially similar to the input.
606
00:45:50.614 --> 00:45:54.276
You don't look at that, but you do have to look at the purpose of the output, right?
607
00:45:54.897 --> 00:45:57.338
So that's where people get confused.
608
00:45:57.478 --> 00:46:02.041
If you look at, is the purpose of the output the same as the purpose of the input?
609
00:46:02.081 --> 00:46:04.162
If so, you're probably not going to win on fair use.
610
00:46:04.546 --> 00:46:05.827
That's very interesting.
611
00:46:05.987 --> 00:46:09.070
You know, where are you thinking that the harm is most visible right now?
612
00:46:09.691 --> 00:46:20.922
I mean, look, if you look at the 120 plus cases that are out there, the vast majority of those cases are in the world of text or book publishing, newspaper publishing.
613
00:46:20.942 --> 00:46:22.383
You know, it's the text world.
614
00:46:22.483 --> 00:46:23.924
And so so that is, I think.
615
00:46:24.905 --> 00:46:47.515
the biggest challenge there are obvious there are a lot of cases in in the music space uh audio visual works movie studios have brought quite a few uh lawsuits as well um but without a doubt the biggest the largest number of cases in text and so i think that's the the most challenging area and for the reasons i gave it's also the most difficult uh for the court to decide as well
616
00:46:47.976 --> 00:46:54.540
Do image models skew closer to music in your mind than the language model?
617
00:46:54.780 --> 00:46:58.123
Because surely creating a moving image is,
618
00:46:59.255 --> 00:47:09.621
I would argue that at least 50% of it is going to be for some form of entertainment, you know, whether you kind of, you know, whether you suggest that, oh, no, it's for scientific diagrams and whatever else.
619
00:47:09.701 --> 00:47:10.841
And sure, I'm sure it would be.
620
00:47:10.862 --> 00:47:18.606
But like there's there's less defensible use cases that differ from what seems to be the primary copyright issue.
621
00:47:19.066 --> 00:47:19.887
Yeah, yeah, yeah.
622
00:47:20.187 --> 00:47:21.548
No, I think that's right.
623
00:47:21.568 --> 00:47:29.854
I think there's a spectrum of types of works and whether they're likely to be infringing or not.
624
00:47:29.894 --> 00:47:37.860
On one end of the spectrum, you have music, which, like I said, I think is absolutely hands down infringing because it's not going to be a fair use.
625
00:47:38.180 --> 00:47:39.862
On the other end of the spectrum, you have text.
626
00:47:40.302 --> 00:47:47.448
which I think a lot of those cases, like I said, especially the newspaper cases, will not be fair use too, but they're at least a little more difficult to handle.
627
00:47:47.548 --> 00:47:48.889
And then you raised images.
628
00:47:49.229 --> 00:47:50.450
Images is closer to music.
629
00:47:50.510 --> 00:48:00.218
It's not as far because you can envision different uses of images that have different purposes, but it's very, very close to the music cases, I think.
630
00:48:00.538 --> 00:48:11.568
Well, I mean, just colloquially in my life, I mean, I'll say that, you know, and maybe this is just kills my music creds or any coolness I may have built up.
631
00:48:12.069 --> 00:48:18.515
But I very happily listen to there's there's one Spotify playlist that's basically like.
632
00:48:19.584 --> 00:48:27.628
you know, Amalfi Coast yacht, you know, like, you know, Amalfi Coast kind of, you know, mid-century music.
633
00:48:28.508 --> 00:48:30.389
And I absolutely love it.
634
00:48:30.770 --> 00:48:35.692
And, you know, I had a whole party at my house where that was what was playing in the background.
635
00:48:35.712 --> 00:48:36.853
And avowedly...
636
00:48:37.714 --> 00:48:40.839
You know, there isn't an artist on the other end of that transaction.
637
00:48:40.859 --> 00:48:47.789
And like you said, it is fully replacing that experience that used to have to be given to me by an artist.
638
00:48:48.030 --> 00:48:53.198
And that's not happening with writing, at least not that I know of, you know, or images.
639
00:48:53.638 --> 00:48:56.839
And even more so because music is largely streamed these days, right?
640
00:48:57.279 --> 00:48:57.959
Yeah.
641
00:48:58.039 --> 00:49:02.520
And there's a, this was a little crude, but a pot of cash, if you will, right?
642
00:49:02.840 --> 00:49:10.362
That the Spotify's of the world distribute to the artists and the songwriters and what have you.
643
00:49:10.782 --> 00:49:21.145
And if all of a sudden AI is taking up that space in what people listen to, less and less money are going to those artists.
644
00:49:22.505 --> 00:49:23.886
Um, and that's a problem.
645
00:49:23.926 --> 00:49:31.769
So getting to my point about music being the easiest cases in my mind is you can show a direct line to the harm too.
646
00:49:31.829 --> 00:49:36.071
You can say, look, we're not getting as much money because AI is out there and people are listening to AI.
647
00:49:36.191 --> 00:49:42.213
And I think, uh, a lot of people, when you open their eyes to this, you're like, this is not a real band.
648
00:49:42.233 --> 00:49:44.054
This is not a real person you're listening to.
649
00:49:44.594 --> 00:49:46.155
Um, we'll get upset about it.
650
00:49:46.415 --> 00:49:48.756
And, but unfortunately, at least right now,
651
00:49:49.656 --> 00:49:52.518
Uh, there does not seem to be any way to stop that from happening.
652
00:49:52.578 --> 00:49:54.879
Like to say, no, I only want to listen to human artists.
653
00:49:55.640 --> 00:49:59.482
Um, uh, let me know when, when some, something is AI.
654
00:49:59.502 --> 00:50:02.364
Uh, I've had long discussions with my friends about this.
655
00:50:02.384 --> 00:50:05.145
Like, how do you tell, I don't want to listen to these, this thing, you know?
656
00:50:05.906 --> 00:50:06.546
And it's very difficult.
657
00:50:06.566 --> 00:50:07.447
And it's very difficult.
658
00:50:07.487 --> 00:50:12.870
So, but, but that just shows still, we are at the very much in the early days of AI and copywriting.
659
00:50:13.010 --> 00:50:18.392
That there's still a lot of challenges that need to be addressed.
660
00:50:19.072 --> 00:50:21.433
We have to figure out a lot, I think, and this is just one area.
661
00:50:21.693 --> 00:50:23.313
How can it possibly keep up?
662
00:50:23.373 --> 00:50:28.115
Because, I mean, I think back to, I mean, it took 20 years at least for...
663
00:50:29.215 --> 00:50:38.863
for people to start legislating against revenge porn, for example, or something you might have shot privately with another individual that you put out on the internet.
664
00:50:39.343 --> 00:50:45.328
That was wrong the entire time it was happening over the 20 years of the internet, yet it was only legislated 20 years later.
665
00:50:45.929 --> 00:50:52.914
With a product and a technology that's moving this fast, what hope do we have?
666
00:50:53.115 --> 00:50:56.397
Is there a mechanism that could be reasonably fast enough
667
00:50:57.398 --> 00:51:04.787
to legislate or or, you know, nanny these things in real time or speed up our responses to them.
668
00:51:05.148 --> 00:51:06.890
So the answer, unfortunately, is no.
669
00:51:06.970 --> 00:51:07.190
Right.
670
00:51:07.331 --> 00:51:14.800
We all know Congress takes a long time to respond and to get any legislation out of Congress is is difficult when it's
671
00:51:15.903 --> 00:51:17.485
not as complex as AI.
672
00:51:17.905 --> 00:51:28.215
And when you don't have a big time players like Google and meta and, and, and, and it's in down the road to deal with.
673
00:51:28.415 --> 00:51:28.635
Right.
674
00:51:28.795 --> 00:51:31.999
So the idea of getting Congress to solve this problem,
675
00:51:32.599 --> 00:51:40.264
or whatever problem, whether it's transparency or fair use or what have you, is I wouldn't bet on that, at least not in the near future.
676
00:51:40.304 --> 00:51:40.445
Right.
677
00:51:40.465 --> 00:51:40.825
That's going to.
678
00:51:41.025 --> 00:51:41.245
Yeah.
679
00:51:41.285 --> 00:51:42.306
Then you look to the courts.
680
00:51:42.326 --> 00:51:44.447
And I think what we're doing is we're looking to the courts.
681
00:51:44.908 --> 00:51:46.749
Now, the courts are not moving all that quickly either.
682
00:51:47.870 --> 00:51:51.492
As I mentioned, there's about 120 cases or more out there.
683
00:51:52.113 --> 00:51:54.735
And we've gotten at this point.
684
00:51:55.475 --> 00:51:57.476
Two district three district court opinions.
685
00:51:58.536 --> 00:52:01.157
We might get an appellate court decision with the next few months.
686
00:52:01.177 --> 00:52:03.097
That's the Thompson Reuters versus Ross case.
687
00:52:03.117 --> 00:52:04.617
So that's a little bit of a different case.
688
00:52:06.038 --> 00:52:11.319
But other than that, these other hundred plus cases that are out there, it's going to take years for them to be decided.
689
00:52:11.479 --> 00:52:14.120
Even when they decided that's only going to be by district court.
690
00:52:14.440 --> 00:52:17.623
They're going to be appealed, no matter which side loses.
691
00:52:17.643 --> 00:52:18.944
They're going to go to appeal.
692
00:52:19.084 --> 00:52:20.686
And so it's going to go to an appellate court.
693
00:52:20.706 --> 00:52:21.947
That will take some more time.
694
00:52:22.347 --> 00:52:24.549
And then eventually, Supreme Court will take it on.
695
00:52:25.170 --> 00:52:29.273
So we are still talking years and years away from resolution here.
696
00:52:29.554 --> 00:52:34.758
And that's what makes this especially difficult, is that there's no immediate solution on the
697
00:52:36.940 --> 00:52:44.687
market-based solutions, which is trying to get AI companies to do the right thing and to license works from the creative community.
698
00:52:45.203 --> 00:52:57.372
I've often said that, you know, when the sort of French Native American War happened, you know, like the Native Americans didn't spend a lot of time making better bows and arrows.
699
00:52:57.412 --> 00:52:58.953
They got their hands on guns.
700
00:52:59.434 --> 00:53:11.443
And I sort of feel that way a little bit about AI, you know, is that the best way to fight back if you're a creator or an artist is to use it and become more prolific.
701
00:53:11.463 --> 00:53:11.563
Yeah.
702
00:53:11.723 --> 00:53:12.484
Output more.
703
00:53:12.584 --> 00:53:13.145
Do more.
704
00:53:13.945 --> 00:53:21.953
You know, this episode is sponsored by the Future of Life Institute, whose focus is on AI that benefits humanity rather than disenfranchising people.
705
00:53:22.314 --> 00:53:30.502
What does it look like for copyright owners, protectors, artists, creatives to use AI in a way that is...
706
00:53:31.583 --> 00:53:36.087
additive and protects themselves and protects the creative community.
707
00:53:36.107 --> 00:53:49.198
Yeah, I think there's a lot of ways that authors and artists and others, creatives, can use AI in a very productive, helpful, respectful, responsible manner.
708
00:53:49.718 --> 00:53:51.680
I think the key is to just remember AI.
709
00:53:52.140 --> 00:53:56.061
That AI is a tool, just like a camera is a tool if you're a photographer, right?
710
00:53:56.081 --> 00:54:00.583
That AI is a tool, but you ultimately are at the controls, right?
711
00:54:00.623 --> 00:54:09.446
You are at the controls to decide kind of what you want to do with your creativity, what you want to create, and then use AI accordingly to help you create.
712
00:54:09.926 --> 00:54:12.687
But don't put AI in the control room.
713
00:54:12.827 --> 00:54:17.411
I think that's, to me at least, other people might disagree or have different views.
714
00:54:17.831 --> 00:54:22.616
I think ultimately that should be the end result if you're a creator.
715
00:54:22.796 --> 00:54:28.381
I don't think you need to shy away from using AI, but do it in a responsible, respectful, ethical way.
716
00:54:29.421 --> 00:54:30.943
where you were still at the helm.
717
00:54:30.983 --> 00:54:35.046
You're the creator, but you use AI to help you.
718
00:54:35.066 --> 00:54:45.394
You mentioned writer's block, and it's a great way, I mean, to use AI for ideation to help you get past that, that it was someplace you're stuck and move forward.
719
00:54:45.774 --> 00:54:47.375
Terrific, terrific use of AI.
720
00:54:47.695 --> 00:54:49.277
But I also feel like part of it,
721
00:54:50.072 --> 00:54:54.034
And just for the experience of being a creator, like, yes, making everything easier is great.
722
00:54:54.074 --> 00:55:06.520
But a lot of the self-worth that comes from finishing a novel comes from the ways you had to battle with your own demons through your own writer's block and your own feelings of insecurity and whatever else.
723
00:55:06.580 --> 00:55:10.082
And, you know, so we do lose something when we lose friction.
724
00:55:10.362 --> 00:55:15.224
And I know that's not a copyright conversation, but it's it's something I think about a lot.
725
00:55:15.524 --> 00:55:17.965
No, you're you're you're you're you're exactly right.
726
00:55:17.985 --> 00:55:30.490
I mean, there's a no matter what you do in life is a amount of satisfaction that you as a human being get right when you had this what seems like an insurmountable challenge or a challenge of any type and you overcome that challenge.
727
00:55:31.110 --> 00:55:40.933
Well, if you had to get AI or something else to assist you to overcome that challenge, maybe you don't feel so great at the end of the day or feel as good, certainly, as you would have.
728
00:55:40.993 --> 00:55:44.294
But I'll stick to copyright because that's what I know.
729
00:55:44.814 --> 00:55:54.437
And I'll leave the creative process up to the creators themselves to figure out what's the kind of right percentage or right amount of use of AI.
730
00:55:54.457 --> 00:55:54.657
Yeah.
731
00:55:54.677 --> 00:55:54.937
Yeah.
732
00:55:55.657 --> 00:55:56.518
Well, fair enough.
733
00:55:56.738 --> 00:55:57.298
I'm sorry.
734
00:55:57.659 --> 00:56:00.141
I tried to pull you out of there for a second, but no.
735
00:56:00.801 --> 00:56:01.602
But thank you.
736
00:56:01.702 --> 00:56:03.944
I mean, Keith, thanks for spending this time with me.
737
00:56:04.024 --> 00:56:05.065
I so appreciate it.
738
00:56:05.105 --> 00:56:16.595
I think that this is one of the most interesting and important conversations to be having, you know, in the face of this technology and by creative people and creative companies.
739
00:56:17.195 --> 00:56:21.579
And, you know, and I so appreciate deep diving in and I've learned so much.
740
00:56:21.639 --> 00:56:23.861
So thank you so much for taking the time to chat with me.
741
00:56:24.001 --> 00:56:25.442
Yeah, well, thank you for having me.
742
00:56:25.542 --> 00:56:26.843
I love talking about the topic.
743
00:56:26.883 --> 00:56:30.325
Happy to, as you can tell, to talk about it.
744
00:56:30.445 --> 00:56:33.727
It is our biggest issue at the Copyright Alliance.
745
00:56:34.428 --> 00:56:41.092
And just to call out to anyone who's listened to your podcast, to go to CopyrightAlliance.org.
746
00:56:41.672 --> 00:56:50.778
On our website, we've got a ton of different educational materials, lots and lots of material about AI and other things about copyright that if you're a creator especially, you might find very, very helpful.
747
00:56:51.138 --> 00:56:51.858
Fantastic.
748
00:56:51.878 --> 00:56:52.478
Well, that's great.
749
00:56:52.519 --> 00:56:55.139
And to be honest, this has been so fun.
750
00:56:55.279 --> 00:57:05.943
I'd love to have you back next year or when some more of these cases shake out and AI is running the government from the Oval Office or whatever.
751
00:57:06.343 --> 00:57:15.589
ever eventually happens, I suspect that this conversation will evolve and will remain incredibly important and incredibly relevant.
752
00:57:15.629 --> 00:57:18.491
So I'd love to have you back when you and I can stay in touch.
753
00:57:18.531 --> 00:57:21.493
And if you go like, I think we're at a threshold moment, we should chat again.
754
00:57:22.413 --> 00:57:23.354
You have an open door.
755
00:57:23.374 --> 00:57:24.174
I'd love to have you back.
756
00:57:24.454 --> 00:57:25.635
Yeah, I appreciate that.
757
00:57:25.675 --> 00:57:26.476
And you're exactly right.
758
00:57:26.516 --> 00:57:28.017
As I said, we're at the early days here.
759
00:57:28.697 --> 00:57:32.799
We could become pen pals, I think, because this this this issue is not going away anytime soon.
760
00:57:33.840 --> 00:57:40.303
And and it's just going to continue as these cases get decided and everyone tries to figure out what does that mean?
761
00:57:40.703 --> 00:57:40.924
Right.
762
00:57:41.024 --> 00:57:43.605
And what does that mean for the next case that's going to come down?
763
00:57:44.145 --> 00:57:49.368
That's that's going to be the absolutely huge challenge that we're all going to face, no matter what side of this equation you're on.
764
00:57:49.608 --> 00:57:51.810
Well, Keith, once again, that was great.
765
00:57:51.871 --> 00:57:52.771
I so appreciate it.
766
00:57:52.792 --> 00:57:53.572
I learned so much.
767
00:57:54.073 --> 00:57:56.916
Thank you for coming on the podcast, and hopefully I'll see you again soon.
768
00:57:56.936 --> 00:57:58.578
And please let's stay pen pals.
769
00:57:58.618 --> 00:57:59.839
Stay in touch.
770
00:58:00.280 --> 00:58:02.222
I love nothing more than to chat about all this.
771
00:58:02.862 --> 00:58:03.363
Absolutely.
772
00:58:03.383 --> 00:58:04.504
Thank you very much for having me.
773
00:58:05.385 --> 00:58:06.045
So that was great.
774
00:58:06.065 --> 00:58:10.728
I had a really good time talking to Keith, and I'm sure you could tell that I got a little overexcited a few times.
775
00:58:10.808 --> 00:58:18.532
I think that what's so great is he unpacks a lot of those really difficult things to understand about copyright.
776
00:58:18.712 --> 00:58:27.717
We've been conflating a lot of things, and what should be shocking, I think, to everybody is, sure, the fair use debate is an interesting one, but as Keith told us,
777
00:58:28.477 --> 00:58:31.739
The moment they digitized a product, they copied it.
778
00:58:31.959 --> 00:58:33.460
And that is something they shouldn't have done.
779
00:58:33.620 --> 00:58:40.364
I'm always in this place where I am not cool with the way these tools got to where they've got.
780
00:58:40.444 --> 00:58:44.987
But I also think that, like, I want to make sure that I and everybody around me stays competitive.
781
00:58:46.068 --> 00:58:48.289
And we got to accept the world as it is sometimes.
782
00:58:48.309 --> 00:58:48.969
Yeah.
783
00:58:50.070 --> 00:58:54.054
And I think his point about music is a really interesting one.
784
00:58:54.555 --> 00:58:57.318
You know, there is the purpose is to replace us.
785
00:58:57.418 --> 00:59:08.129
And I think that that's the interesting discussion here is there's the data in, there's the output, but then there's also the fact that this stuff is competing with us.
786
00:59:09.183 --> 00:59:12.965
And we have a right to not be competed against by our own work.
787
00:59:13.105 --> 00:59:23.171
And I think that when we talk about all of the things that are happening and all of the job displacement and things like that, people whose jobs are being displaced are beyond copyright.
788
00:59:23.431 --> 00:59:25.252
The people and the roles...
789
00:59:26.334 --> 00:59:32.475
over years that have done this work, and that work is being used to replace them.
790
00:59:33.295 --> 00:59:36.576
And that's quite unacceptable.
791
00:59:37.136 --> 00:59:38.156
So it's very interesting.
792
00:59:38.936 --> 00:59:40.237
I'm really glad we had that chat.
793
00:59:40.257 --> 00:59:41.417
I might have Keith back.
794
00:59:42.317 --> 00:59:50.779
I'll probably re-listen to and re-digest that and come back a little bit less flustered because I was so excited by the chat.
795
00:59:51.219 --> 00:59:52.499
It sometimes got away from me.
796
00:59:52.559 --> 00:59:54.662
Anyway, thank you for listening to that.
797
00:59:54.722 --> 00:59:56.946
Again, that was sponsored by the Future of Life Institute.
798
00:59:56.966 --> 00:59:59.169
We're incredibly happy to have them as a partner.
799
00:59:59.650 --> 01:00:03.956
Next week, we have Ryan Black and Kit Janae.
800
01:00:04.257 --> 01:00:05.559
They're from Luma Labs.
801
01:00:05.979 --> 01:00:07.962
Luma has created one of the best...
802
01:00:08.523 --> 01:00:13.525
vision models or models for creating cinematic work.
803
01:00:14.006 --> 01:00:22.790
They are a fantastic and impressive company that has, I think the best video model on the market.
804
01:00:23.770 --> 01:00:35.076
And what's interesting is that they've taken a playbook from, from some of the big technology companies like Palantir and kit is what's called a Ford deployed creative.
805
01:00:35.096 --> 01:00:35.176
Wow.
806
01:00:35.836 --> 01:00:44.121
And I think that's a really interesting thing to talk about is the notion that they have created a tool that is incredibly valuable for creative companies.
807
01:00:44.301 --> 01:00:48.003
But they also recognize that it's difficult to use and people are using lots of different tools.
808
01:00:48.443 --> 01:00:55.347
So they have a team of four deployed creatives, people who will work with you to teach you on the models.
809
01:00:55.367 --> 01:00:56.647
And I think that's very interesting.
810
01:00:57.668 --> 01:00:59.189
Anyway, please tune in for that.
811
01:00:59.289 --> 01:01:01.350
Thank you so much for listening and we'll see you next week.